On April 8, 2026, a “dark horse” emerged in the AI video generation field. An open-source AI video generation model called Happy Horse 1.0, without any warm-up, launch event, or official announcement, suddenly appeared on the Video Arena leaderboard of the global authority AI benchmark platform Artificial Analysis, securing the top spot in both text-to-video and image-to-video benchmarks, outperforming mainstream models like ByteDance’s Seedance 2.0, Kuaishou’s Kling 3.0, and Google’s Veo 2.6, sending shockwaves through the industry.
This new model, dubbed “Happy Horse” by Chinese netizens, was developed by Future Light Lab under Alibaba’s Taobao Tmall Group. It uses a unified Transformer architecture with 15 billion parameters and can natively generate video footage along with sound effects, dialogue, ambient sounds, and background music simultaneously. Notably, Happy Horse 1.0 is released under a fully open-source model with commercial licensing, meaning open-source models are for the first time directly competing with closed-source flagship models in video generation capabilities.
Completely dominates authoritative rankings
According to blind test results from Artificial Analysis using the ELO rating system, Happy Horse 1.0’s performance has been nothing short of “terrifying.” In the text-to-video (no audio) category, the model ranked first with 1333 points, surpassing all mainstream models including Seedance 2.0, Keling 3.0, and Veo 2.6. In the image-to-video (no audio) category, it dominated with 1392 points, leaving the second-place finisher more than 50 points behind. In the ELO rating system, a 50-point gap already represents “a generation” of difference.
Even in the two audio-enabled evaluations, Happy Horse ranked second by only a narrow margin. In text-to-video (with audio), it scored 1205 points, just a few points behind Seedance 2.0, while image-to-video (with audio) was only 1 point behind the leader. This demonstrates that the model is a true “all-rounder” rather than a one-trick pony.
Technical Breakthrough: Native Audio-Video Sync and Unified Architecture
Happy Horse’s technical core relies on two key advantages. The first is its native audio-video synchronized generation capability. Most current AI video models use a separate workflow—first generating the video, then independently generating the audio. Happy Horse, however, employs a unified Transformer architecture that can collaboratively generate both video and audio within the same model, naturally aligned without the need for post-production dubbing or mixing.
Next is the unified architecture design with 15 billion parameters. In traditional approaches, video and audio generation are handled by two separate models and then spliced together, whereas the unified architecture allows both to be generated jointly in the same process, greatly improving audio-visual synchronization. This model supports lip-sync in 7 languages, with output quality reaching cinematic standards of 720P/1080P, and video duration ranging from 3 to 45 seconds, supporting both landscape and portrait orientations.
Open Source Strategy Shakes Up the Industry
What shocked the industry most was Happy Horse’s open-source strategy. All model selections are completely open-sourced with commercial licensing, and all model weights, distilled models, super-resolution modules, and inference code are GitHub was publicly released. This contrasts sharply with the recent trend where most AI video models opt for closed-source approaches and charge API fees. However, this model requires at least an H100 to run, making it inaccessible for regular consumer-grade graphics cards.
After the announcement, Alibaba’s Hong Kong-listed shares (9988.HK) surged as much as 8% to 127.9 HKD. Citing sources, Wall StreetCN reported that the model development team was led by Zhang Di, former Vice President of Kuaishou. Initially released quietly under the name “Anonymous Asian AI Lab,” it bypassed pre-launch hype and press conferences, going straight to the top of the rankings—a true Cinderella story in the AI video generation赛道.
Tested head-to-head: each has its own strengths
Based on real-world tests conducted by the designer community, Seedance 2.0 delivers smoother and more natural motion in real human gait scenarios; in physics-based scenarios, Seedance 2.0 offers more solid force feedback and camera pacing; however, in animation-style scenarios, Happy Horse clearly excels in hair movement details, expression liveliness, and scene atmosphere.
Artificial Analysis 榜单出现一匹超越seedance2.0的黑马：HappyHorse-1.0（据说是来自阿里的）
目前在文本和图像转视频（无音频）中排名第一
在文本和图像转视频（有音频）中排名第二
可以去官网体验一下：https://t.co/chxCu8G0qt pic.twitter.com/6QjDALxkVZ
— Smartpig (@Smartpigai) April 8, 2026
Did Google just drop Veo 4?
A new anonymous video model is currently leading both text-to-video and image-to-video on the Artificial Analysis leaderboards.
It’s called HappyHorse-1.0, and it’s very promising.
More examples below. pic.twitter.com/8UYzS55ZaY
— Angry Tom (@AngryTomtweets) April 7, 2026
HappyHorse1.0会超越Seedance2.0吗？？
对比视频来了，个人感觉Seedance2.0略胜一筹，
具体的说不清楚，总觉得更加自然和真实，或许在物理逻辑上和镜头运动上调教的更好。当然，如果HappyHorse不排队，那它更强！！！ pic.twitter.com/OO0BJzBUlT
— 沐阳 (@yyyole) April 8, 2026
HappyHorse-1.0 is now confirmed as Alibaba’s latest video model.
It supports Text-to-Video and Image-to-Video with and without native audio and is already ranking #1 on Video Arena.
API access is planned for launch on April 30.pic.twitter.com/FYw3il5SZq https://t.co/UA9BCV1NDY
— Angry Tom (@AngryTomtweets) April 10, 2026
Industry analysts point out that Happy Horse’s defining characteristic is “aggressive pushing”: excelling at using visual impact to instantly grab viewers’ attention; while Seedance “knows how to pull back” better, more skilled at controlling details and maintaining stability—not necessarily the most stunning, but unlikely to fail. For creators on tight budgets and tight schedules, Happy Horse’s free open-source nature is highly attractive; for product demo scenarios requiring precise physics simulation, Seedance 2.0 is the more reliable choice.
Application Scenarios and Industry Impact
The launch of Happy Horse will have a profound impact on multiple industries. In e-commerce and marketing, it can quickly generate high-quality product showcase videos and personalized advertising shorts, significantly reducing content production costs; in film and animation creation, it can assist with script visualization, generate specific style storyboards, and create VFX scenes, markedly improving efficiency and shortening production cycles; in social media and personal creation, it empowers every user to become a “super creator,” transforming text and image ideas into vivid videos with one click; in education and enterprise services, it generates virtual experiment processes, historical scene recreations, product feature explanation videos, and more, providing more intuitive and efficient information delivery. Now Happy Horse also hasOpen for trialYou can log in and make a 5-second low-resolution video. Feel free to try it if you’re interested. From my own testing, I found the character consistency is worse than Seedance.
Source: KOCPC Chinese